Study On Main Control Factors And Production Prediction Of Single Well Production Of Coalbed Methane Based On Machine Learning

CBM data has many characteristics,such as large amount of data,different types and types of information,redundant information,missing data and so on.The traditional statistics and comparison of the qualitative analysis method has been difficult to quickly find useful information,and based on the processing method of machine learning,data modeling algorithm is efficient,scientific and accurate data feature extraction,the value of data mining.At the same time,this method also provides a new idea for yield control,diagnosis and production prediction in oil and gas industry.In this paper,through literature research,combined with field data,to find out the influence of coal gas production including geology,drilling,fracturing and drainage in 4 aspects of the 47 factors,these data were collected using machine learning data preprocessing method,the qualitative variables of missing data virtualization,KNN interpolation,drawing analysis of factors affecting the relationship between yield and point matrix of qualitative scattered data standardization to eliminate dimensional principal component feature extraction,the 47 factors are compressed into 21 principal components.After the data pretreatment,the K samples are clustered into 3 categories.According to the magnitude of the stable gas flow and the size of the cluster center in each type of wells,these three types of wells are defined as well,medium and inferior wells.After each wells annotation data sets will be completed,80% for training the model,20% for testing the performance of the model and classification model training algorithm using KNN,the model is applied to the test dataset,the accuracy can reach 100%.So far,the main control factor model of CBM production is constructed.Respectively using RPROP neural network,support vector machine and establish multi stable gas production with 21 principal component factors model of quantitative relationship between the stepwise regression model,the model is applied to the test data,the results show that the prediction accuracy of RPROP neural network in more than 85%,the normalized mean absolute error is only 0.04.Therefore,the model is optimized to build a prediction model of stable and daily output of coalbed methane.The application of the main controlling factors of coalbed methane production and methane yield diagnosis model of gas production prediction model for diagnosis and prediction of 4 wells,4 wells have predicted a well as inferior wells,wells and wells.By comparing the characteristic parameters of the difference,find out the cause of the low yield wells main control is see pneumatic liquid level and liquid drop rate.So as to indicate the direction of the improvement of the engineering parameters of the well.